Currently, virtualization technology has been widely used. Virtualization technology provides an approach capable of accommodating a plurality of Virtual Machines (VMs) on a large server, which reduces total cost of user and improves reliability and flexibility of applications, thereby reducing possibility of providing excessive computing resource by user for spike of various workloads. Moreover, virtualization technology helps to isolate important applications, such that different applications running on different virtual machines will not share operation system, thus critical failure caused by an application will not affect applications in any other virtual machines. The virtualization technology is an important and fundamental technology in “cloud computing” technology, and since it has the capabilities of on-demand configuration and dynamic migration, it becomes a key for realizing dynamic and scalable extension of cloud computing. In order to realize the high flexibility of the cloud computing to satisfy policies and requirements of services or management, dynamic migration and aggregation of virtual machines among physical machines need to be performed frequently, so as to achieve the objectives of reducing energy consumption, improving resource utilization and lowering management costs, etc.
However, virtual machine migration is a highly I/O sensitive process. During the process that the virtual machine is migrated, since the services or applications thereon are still in operation, this will produce unexpected influence on application clients of the virtual machine. For example, during the process of virtual machine migration, an application client of the virtual machine issues a request to the virtual machine, this will cause a response failure, thus a retransmission request needs to be sent, the criterion of TCP protocol for controlling whether a data segmented needs to be retransmitted is to set a retransmission timer. A retransmission timer is started when a data segment is sent. If an acknowledgement is received before the timer is timed out, the retransmission time is turned off. If no acknowledge is received before the timer is timed out, the data segment is retransmitted. During the process of selecting retransmission time, TCP has to be adaptive. It needs to give a suitable data retransmission time based on communication condition of Internet at that time. Currently, a frequently used algorithm is a dynamic algorithm that continuously adjust time-out time interval. Its operation principle is as follows: a variable Round Trip Time (RTT) is maintained for each connected TCP, which is used to store an estimated value closest to time required for a round trip between current and destination end. If no acknowledge is received before the timer is timed out, value of RTT is doubled. As number of times of connection failure increases, value of RTT is larger and larger. During the process of virtual machine migration, since value of RTT is small, number of times of retransmission will increase. If the virtual machine migration has already been completed, since value of RTT is larger and larger as number of times of retransmission increases, response delay will occur at client.
Thus, when a virtual machine is being migrated, a suitable retransmission time needs to be set for application client, so as to respond to request of the client in time.